AI Engineer · Amman, Jordan

Sami
Kayyal.

I build retrieval systems

Data Science & AI grad who ships LLM systems end to end — retrieval, evaluation, backend, deployment. I like turning messy archives and messier data into things people can actually use.

Location
Amman, Jordan
Degree
PSUT — DS & AI, 2026
Languages
Arabic · English
// about

I'm an AI engineer who'd rather ship the whole thing than hand off a notebook. Lately that's meant an agentic RAG platform over a 27k-file archive, a multilingual analysis pipeline that cut manual review ~70%, and leading a small team from model logic to product decisions.

I care about the unglamorous parts — retrieval that actually retrieves, evaluation you can trust, and deployments that survive real traffic. Give me a vague problem and a pile of data and I'm happy.

Selected work

02 — featured

Where I've worked

Jun 2026 — Present

Data Science & AI Intern · Mindset, Amman

Built a multilingual survey-analysis pipeline for Arabic (incl. dialectal) and English. Automated dedup, language-preserving embeddings, PCA and HDBSCAN clustering — cutting manual review time ~70%.

Feb — May 2026

AI Intern · Mocion, UAE (Remote)

Proposed and led a 3-engineer project to predict player engagement and no-show risk. Shipped three predictive workflows for event viability, campaign targeting, and forecasting — then handed off model logic and findings to stakeholders.

Dec 2024 — Mar 2025

Software Developer · Gravity Syria, Damascus (Remote)

Built a Flutter check-in & subscription system for a trampoline park — pricing tiers, expiry alerts, a lightweight inventory module, and an ops analytics dashboard.

🎓 education BSc Data Science & Artificial Intelligence · Princess Sumaya University for Technology · 2026 · GPA 80% (Very Good)
// the stack
RAG·Agentic systems·Hybrid retrieval·Embeddings·Python·FastAPI·Docker·Cloud Run·scikit-learn·TensorFlow·OpenCV·Flutter·
// say hi

Let's build
something.

Got a problem buried in data, or an LLM idea you want shipped for real? I'm around.

CASE STUDY · LIVE PROJECT

UniRAG

Agentic RAG over a 27,000+ file university course archive — source-grounded answers, deployed for real traffic.

overview

UniRAG is a production-grade agentic RAG platform built to make a 27,000+ file university archive genuinely searchable. It plans each query, combines metadata-aware keyword and semantic retrieval across course documents, slides, notebooks, code, transcripts, and datasets, then returns a source-grounded answer with the evidence attached.

stack
FastAPI ChromaDB SQLite FTS5 Docker Google Cloud Run Firestore
what I built
01

Engineered a production RAG platform with query planning and keyword-semantic hybrid retrieval across structured metadata, SQLite FTS5, and ChromaDB.

02

Deployed the FastAPI service on Google Cloud Run with Docker, HTTPS, health/readiness checks, and Secret Manager credentials.

03

Hardened it for real traffic: Firestore-backed quotas, signed sessions, Turnstile protection, bounded concurrency, and request-cancellation handling.

try it live ↗
CASE STUDY · GRADUATION PROJECT

FitnessCam

Computer vision, live coaching, and Arabic AI planning combined in one production-oriented mobile platform.

overview

FitnessCam is an end-to-end fitness platform built around three AI capabilities: photo-based body-composition analysis, live camera exercise coaching, and personalized nutrition and workout planning. I built the Flutter client, FastAPI backend, inference pipelines, persistence layer, and Google Cloud deployment as one integrated product.

stack
Flutter FastAPI PyTorch YOLOv8 Pose MediaPipe OpenCV PostgreSQL Google Cloud Run
what I built
01

Built the complete Flutter mobile experience across authentication, onboarding, profiles, progress tracking, body-composition check-ins, and personalized plans.

02

Developed the computer-vision pipelines: MediaPipe segmentation and dual-view body measurement, plus YOLOv8 Pose and temporal analyzers for live squat and biceps-curl feedback.

03

Added Jordanian Arabic AI coaching that turns user goals, biometrics, preferences, and constraints into editable weekly nutrition and workout plans.

04

Productionized the FastAPI backend with Docker, Cloud Run, Cloud SQL, Google Cloud Storage model hydration, Secret Manager, migrations, and readiness checks.

view the repository ↗